Investor Demo USA

This Milestone will advance the POC and make MLReef presentable to an audience of investors and other interested people. Depending on the level of maturity maybe also already the first very friendly alpha users

Board

Tech-demo presentation sequence

This is the ideal demonstration sequence in which we will do the live demo show of MLReef. This list also contains possible shortcuts we can take. Shortcuts are the exclusion of functionality which we assume to be very work-intensive.

Shortcuts: The login credentials can be hardcoded user:camillo password:password

2. ✅ My Projects Page

Initially, the user will not have private projects. We will then switch to the list of public projects (tab) and click on the "Ham10000" project.

Shortcuts: The list of public projects can be hardcoded in the GUI

3. ✅ Project overview of "Ham10000"

Show the data set (repo overview). Show the file structure (navigate into the folders) and the README.md at the bottom.

Shortcuts: none

4. ✅ Fork project

Fork the Ham10000 project into the user's workspace

Shortcuts: none

5. ✅ Jump to own project fork of Ham10000.

Use either the navigation or the Projects Overview Page to navigate to our own fork of the Ham10000 repository.

Shortcuts: none

6. ✅ Clone project locally

Switch to the local command line and clone our private version of Ham10000. The correct clone-link has to be displayed in the GUI so that it can be copied.

Add some files locally and push commit back to the server.

Shortcuts: none

7. ❌ Show that file-counter on the folder has changed.

Shortcuts: none

8. ✅👽Show commit-history with changes made.

This step is optional

Shortcuts: none

9. ✅ Process the Data with a Data Pipeline

Open the Data Pipeline View and:

select the correct input data
add operation augment data (5x)
add operation random crop (px x px)
add operation rotate (degrees)

Shortcuts: none

10. ✅ Execute Data Pipeline

It needs to generate the data and save the data as a complete data instance.

Shortcuts: none

Issues: Save and Execute Data Pipeline (CI/CD Pipeline)

11. ❌ Show In-Progress Pipeline

Open the Data Instances View which shows the data pipeline in progress.

Shortcuts: none

Issue: Pipeline needs to communicate to frontend view that pipeline is in progress. Therefore also that it has finished.

Issue: Output of data pipeline needs to be saved and stored outside the docker image - S3 bucket?

Issue: Data file needs to be name filename_hashpipeline

12. ❌👽 Merge Finished Data Pipeline

When the pipeline is done, show the Data Instances View, with the new active data instance. Save the active data instance (which will merge it into the master branch.

Shortcuts: The data instance can be merged into master directly by the pipeline - without manually saving it and completely without the Data Instances View

13. ✅ Show Experiments View

Open the experiments view which initially will be empty.

Shortcuts: none

14. ✅ Create new experiment

Select data (from data instance or new data branch from data pipeline)
Select algorithm (e.g. Resnet-50 with default params)

Shortcuts: none

15. ✅ Run Experiment

Opens modal view where the user selects training epochs and the size of the VM (CPU, RAM, GPU)

Shortcuts: Epochs can be hardcoded, VM size selection can be completely fake

16. ✅ Show Experiments Overview

With a running Experiment. When the experiment is done, show the Expanded View in Experiment Overview with the executive summary.

Shortcuts: none

17. ❌ 👽 Model Inference

This step is optional Tensor Board (from Tensor Flow) gives us PNG representation of the most important graphs

In Scope

Multy Tenancy
Editing of files via GUI
GPU Integration for Runners
Hosting on MLReef Infra Structure

Out of Scope

Real Data Instances

Legend

❌ … This step currently has to be skipped

🚧 … related tasks are in progress

✅ … This Step can already be demonstrated

👽 … This step is optional

Strech Goals

self-hosting Gitlab on our Infrastructure
real login
✅ pipeline scaling via cloud-providers